Latest AI News
We’re feeling cynical about xAI’s big deal with Anthropic
Anthropic and xAIannounced a big partnershipthis week, with Anthropic buying all the compute capacity at xAI’s Colossus 1 data center in Tennessee. On the latest episode ofTechCrunch’s Equity podcast, Kirsten Korosec, Sean O’Kane, and I discussed what the deal might mean for xAI’s parent company SpaceX, as SpaceX prepares to go public andapparently plans to dissolve xAIas a separate organization. Kirsten did her best to offer “a positive view” on the partnership — after all, it’s a new way for xAI to make money. But she also noted that this also suggests xAI isn’t doing much when it comes to training its own frontier AI models, and it’s harder for the company to position itself as a “forward-looking, innovative” business when that’s the case. Then Sean asked: “Why be positive when you can be cynical?” In his view, this seems like “a major heat check before the IPO.” Yes,becoming a neocloudmight be “a more believable business in the near term,” but it’s less likely to get outside investors excited in the long term. (And then there’sthe environmental lawsuitthat xAI is facing over Colossus 1.) Keep reading for a preview of our conversation, edited for length and clarity. Sean O’Kane:I always love a surprise, especially when everybody’s eyes [are] on another ball,a major trialthat’s happening. Seemingly out of nowhere this week, SpaceX and therefore its AI subsidiary xAI — which apparently no longer exists now, or is imminently not about to exist, which we can get to — struck a deal with Anthropic. Basically, the real version of the deal is that Anthropic’s essentially taking over all of the compute at the data center known as Colossus 1 in Memphis, Tennessee, to focus on Anthropic’s more enterprise-focused AI products. There’s been a lot of reporting about how [Anthropic’s] been looking for more compute […] and it seems like an escape valve for them to be able to strike this deal and get access to all this compute. In the near term, for xAI and for SpaceX, yes, they are a neocloud now, in the sense that they had to do something with all this compute that they were building, because it certainly seems like they were not going to need it for Grok — which, outside of X, is not burning up the world as far as becoming the new hot consumer chat bot. Kirsten Korosec:And we should say that in terms of what a neocloud is, for those who don’t know, this is the idea of buying GPUs from Nvidia and the like, and renting those out as opposed to using those for their own AI, training their own AI models. So this is a different kind of business, andthe point that our AI editor, Russell Brandom, makesis that a lot of companies are building out data centers, but if given a choice between, do they rent them out [or using them to train their own models], they are still prioritizing using this compute for their own internal AI model training. I think that’s an important point and one that suggests that maybe xAI isn’t doing so much on the AI model training [side] Anthony Ha:Right, and as Sean was alluding to, most people would not necessarily think of Grok as — not only that it’s known for some pretty unpleasant, if notdownright illegal, content, but also it’s not necessarily super cutting edge. Especially if we start talking about enterprise AI, which I know we’re gonna be getting into later in this episode, you don’t hear a lot about people using Grok for work-critical tasks. And so the question becomes: How can xAI actually make money? And apparently just selling the infrastructure could be one of the main ways to do it. Kirsten:And you could take a positive view on that, right? They figured out a way to make money. But I think that when you are positioning your company — in this case, SpaceX-slash-xAI — as a forward-looking, innovative company, that’s tougher to sell if you are simply just renting out your GPUs and not using them for that innovation. Sean:But why be positive when you can be cynical? Which is to say that this seems like a major heat check before the IPO that we’re about to see get rammed into the markets with SpaceX. Anthony, you mentioned not only is Grok not being used for big enterprise tasks, there’s been reporting that xAI employees wereusing other models, they weren’t even using [Grok] internally, and that caused this big shakeup inside of xAI, postacquisition from SpaceX, that involved essentiallyall the co-founders leaving other than Elon Musk, [and] him basically saying he’s starting from scratch on xAI, despite the fact that SpaceX paid $250 billion for it in the run up to this mega-IPO. And now he’s saying thatthey’re going to dissolve xAIas a separate entity inside SpaceX altogether. He’s starting to call the whole thing SpaceXAI, because this man loves nothing but to ruin a brand that has some value to it — see Twitter. This may be a more believable business in the near term, and so on some level, I could see this being maybe more attractive to investors come IPO time, because it’s like a bit more reliable and certainly more real than them being a frontier lab developer. But it’s also not the kind of business that’s going to draw the same — at least, in a normal environment — outside investment that we’re seeing go into all the frontier labs. That’s maybe one of the biggest tension points we’ve seen develop during this IPO process. Loading the player…
View
‘We Have Swarms of Agents’: Yasmeen Ahmad on Google’s Future of Enterprise AI
Google has introduced Knowledge Catalog, a context engine to enhance data interpretation in multi-cloud environments.
View
How to Use Netflix's New AI Voice Search Feature: A Step-by-Step Guide
Netflix recently began rolling out a new way for viewers to search for shows and movies on its platform. While we can search for content online via voice dictation, it merely presents results based on keywords. However, the new native AI-based voice search tool will provide contextual search results, taking the intent of the user's query into account. Currently available to a small set of users in beta, the content streaming company is asking users to test the new functionality and provide feedback on how it can be refined, while also pointing out the bugs and issues. The company has yet to announce when the stable version of the AI search tool will be rolled out to a wider global user base.
View
Voice AI in India is hard. Wispr Flow is betting on it anyway.
India’s internet users already rely heavily on voice notes, voice search, and multilingual messaging. Turning those habits into a scalable AI business, however, remains difficult because of the country’s linguistic complexity, mixed-language usage, and uneven monetization patterns.Wispr Flowis betting the opportunity is worth the challenge. The Bay Area-headquartered startup, which builds AI-powered voice input software, says India is now its fastest-growing market, even though voice-based AI products remain early and fragmented in the South Asian nation. That growth has pushed Wispr Flow to expand more aggressively for Indian users,beginning with Hinglish— a hybrid mix of Hindi and English commonly spoken by locals. The startup is also planning broader multilingual voice support, a local hiring push, and, eventually, lower pricing as it looks to expand beyond white-collar users and into Indian households. Earlier waves of voice technology in India —from digital assistantstoWhatsApp voice notes— largely revolved around convenience. AI startups such as Wispr Flow are now betting that generative AI can turn those habits into a broader computing layer. To make the product more relevant for Indian users, Wispr Flow began beta testing a Hinglish voice model earlier this year andlaunched on Android— India’sdominant mobile operating system— after initially debuting on Mac and Windows beforeexpanding to iOSin 2025. Co-founder and CEO Tanay Kothari told TechCrunch that the startup initially saw adoption in India largely among white-collar professionals such as managers and engineers, but it’s increasingly seeing broader usage patterns emerge, including among students and older users being onboarded by younger family members. India has emerged as Wispr Flow’s second-largest market after the U.S. in terms of both users and revenue, Kothari said, with growth accelerating following the startup’s recent India-focused push. The startup has seen faster growth following the rollout of Hinglish support, benefiting from the widespread habit among Indian users of mixing Hindi and English in everyday conversations, particularly as users began expanding beyond work-focused use cases into more personal communication. “The biggest thing is people are starting to use it more in personal apps,” Kothari said, pointing to messaging platforms such as WhatsApp and social media apps where users frequently switch between Hindi and English while speaking. Wispr Flow, Kothari said, was growing about 60% month over month in India earlier this year, but growth accelerated to around 100% following its recent India launch campaign. The startup last month rolled out abroader marketing pushin the country, including a launch video from Kothari and offline campaigns in Bengaluru aimed at introducing the product to more mainstream users. Kothari told TechCrunch that Wispr Flow plans to expand its multilingual voice support over the next 12 months, allowing users to switch between English and other Indian languages beyond Hindi while speaking. In December, the startupintroduced India-specific pricingat ₹320 (around $3.4) per month for annual plans, significantly lower than its standard $12 monthly pricing globally. The startup eventually wants to bring costs down even further — potentially to as low as ₹10–20 (around 10–20 cents) per month — as it looks to expand beyond white-collar and urban users. “I want every single person in the country to be able to use Wispr Flow, and that’s what we’re really building for,” Kothari said. “That’s going to happen slowly and steadily.” Earlier this year, Wispr Flow hired Nimisha Mehta to lead its India operations as it looks to expand its local presence. Kothari told TechCrunch the startup plans to grow to around 30 employees in India over the next year, building out consumer growth, partnerships, and enterprise teams alongside existing engineering and support functions. The startup currently has about 60 employees globally. Wispr Flow is not alone in viewing India as a key market for voice-based AI products. Companies including ElevenLabs have highlighted India as animportant growth marketforsome time. Similarly, local startups such as Gnani.ai, Smallest AI, and Bolna havecontinued attracting investor interestas voice-based AI tools gain wider adoption across consumer and business use cases. Nevertheless, turning voice AI into a mainstream consumer product in India remains challenging despite growing interest from startups and investors. “India is the ultimate stress test for voice AI,” Neil Shah, vice president of research at Counterpoint Research, told TechCrunch, adding that “linguistic, accent, and contextual friction” continue to slow wider adoption. Data shared with TechCrunch from Sensor Tower shows Wispr Flow was downloaded more than 2.5 million times globally between October 2025 and April 2026, with India accounting for 14% of installs during the period, making India its second-largest market by downloads (after, as mentioned, the U.S.). India, however, contributed only around 2% of Wispr Flow’s in-app purchase revenue during the same period, according to Sensor Tower. However, the startup remains largely desktop-driven globally. Wispr Flow’s usage in India, Kothari said, is currently split roughly 50:50 between desktop and mobile, compared with an 80:20 desktop-heavy mix in the U.S. Kothari said Wispr Flow sees strong repeat usage among its users, claiming roughly 70% retention after 12 months globally and in India. Moreover, the startup currently employs two full-time linguistics PhDs as it continues refining multilingual voice models and expanding support for additional Indian language combinations.
View
So you’ve heard these AI terms and nodded along; let’s fix that
Artificial intelligence is changing the world, and simultaneously inventing a whole new language to describe how it’s doing it. Spend five minutes reading about AI and you’ll run into LLMs, RAG, RLHF, and a dozen other terms that can make even very smart people in the tech world feel insecure. This glossary is our attempt to fix that. We update it regularly as the field evolves, so consider it a living document, much like the AI systems it describes. Artificial general intelligence, or AGI, is a nebulous term. But it generally refers to AI that’s more capable than the average human at many, if not most, tasks. OpenAI CEO Sam Altman once described AGI as the “equivalent of a median human that you couldhire as a co-worker.” Meanwhile,OpenAI’s charterdefines AGI as “highly autonomous systems that outperform humans at most economically valuable work.” Google DeepMind’s understanding differs slightly from these two definitions; the lab views AGI as “AI that’s at least as capable as humans at most cognitive tasks.” Confused? Not to worry —so are experts at the forefront of AI research. An AI agent refers to a tool that uses AI technologies to perform a series of tasks on your behalf — beyond what a more basic AI chatbot could do — such as filing expenses, booking tickets or a table at a restaurant, or even writing and maintaining code. However, as we’veexplained before, there are lots of moving pieces in this emergent space, so “AI agent” might mean different things to different people. Infrastructure is also still being built out to deliver on its envisaged capabilities. But the basic concept implies an autonomous system that may draw on multiple AI systems to carry out multistep tasks. Think of API endpoints as “buttons” on the back of a piece of software that other programs can press to make it do things. Developers use these interfaces to build integrations — for example, allowing one application to pull data from another, or enabling an AI agent to control third-party services directly without a human manually operating each interface. Most smart home devices and connected platforms have these hidden buttons available, even if ordinary users never see or interact with them. As AI agents grow more capable, they are increasingly able to find and use these endpoints on their own, opening up powerful — and sometimes unexpected — possibilities for automation. Given a simple question, a human brain can answer without even thinking too much about it — things like “which animal is taller, a giraffe or a cat?” But in many cases, you often need a pen and paper to come up with the right answer because there are intermediary steps. For instance, if a farmer has chickens and cows, and together they have 40 heads and 120 legs, you might need to write down a simple equation to come up with the answer (20 chickens and 20 cows). In an AI context, chain-of-thought reasoning for large language models means breaking down a problem into smaller, intermediate steps to improve the quality of the end result. It usually takes longer to get an answer, but the answer is more likely to be correct, especially in a logic or coding context. Reasoning models are developed from traditional large language models and optimized for chain-of-thought thinking thanks to reinforcement learning. (See:Large language model) This is a more specific concept that an “AI agent,” which means a program that can take actions on its own, step by step, to complete a goal. A coding agent is a specialized version applied to software development. Rather than simply suggesting code for a human to review and paste in, a coding agent can write, test, and debug code autonomously, handling the kind of iterative, trial-and-error work that typically consumes a developer’s day. These agents can operate across entire codebases, spotting bugs, running tests, and pushing fixes with minimal human oversight. Think of it like hiring a very fast intern who never sleeps and never loses focus — though, as with any intern, a human still needs to review the work. Although somewhat of a multivalent term, compute generally refers to the vitalcomputational powerthat allows AI models to operate. This type of processing fuels the AI industry, giving it the ability to train and deploy its powerful models. The term is often a shorthand for the kinds of hardware that provides the computational power — things like GPUs, CPUs, TPUs, and other forms of infrastructure that form the bedrock of the modern AI industry. A subset of self-improving machine learning in which AI algorithms are designed with a multi-layered, artificial neural network (ANN) structure. This allows them to make more complex correlations compared to simpler machine learning-based systems, such as linear models or decision trees. The structure of deep learning algorithms draws inspiration from the interconnected pathways of neurons in the human brain. Deep learning AI models are able to identify important characteristics in data themselves, rather than requiring human engineers to define these features. The structure also supports algorithms that can learn from errors and, through a process of repetition and adjustment, improve their own outputs. However, deep learning systems require a lot of data points to yield good results (millions or more). They also typically take longer to train compared to simpler machine learning algorithms — so development costs tend to be higher. (See:Neural network) Diffusion is the tech at the heart of many art-, music-, and text-generating AI models. Inspired by physics,diffusion systems slowly “destroy” the structure of data— for example, photos, songs, and so on — by adding noise until there’s nothing left. In physics, diffusion is spontaneous and irreversible — sugar diffused in coffee can’t be restored to cube form. But diffusion systems in AI aim to learn a sort of “reverse diffusion” process to restore the destroyed data, gaining the ability to recover the data from noise. Distillation is a technique used to extract knowledge from a large AI model with a ‘teacher-student’ model. Developers send requests to a teacher model and record the outputs. Answers are sometimes compared with a dataset to see how accurate they are. These outputs are then used to train the student model, which is trained to approximate the teacher’s behavior. Distillation can be used to create a smaller, more efficient model based on a larger model with a minimal distillation loss. This is likely how OpenAI developed GPT-4 Turbo, a faster version of GPT-4. While all AI companies use distillation internally, it may have also been used by some AI companies to catch up with frontier models. Distillation from a competitor usuallyviolatesthe terms of service of AI API and chat assistants. This refers to the further training of an AI model to optimize performance for a more specific task or area than was previously a focal point of its training — typically by feeding in new, specialized (i.e., task-oriented) data. Many AI startups are taking large language models as a starting point to build a commercial product but are vying to amp up utility for a target sector or task by supplementing earlier training cycles with fine-tuning based on their own domain-specific knowledge and expertise. (See:Large language model [LLM]) A GAN, or Generative Adversarial Network, is a type of machine learning framework that underpins some important developments in generative AI when it comes to producing realistic data — including (but not only) deepfake tools. GANs involve the use of a pair of neural networks, one of which draws on its training data to generate an output that is passed to the other model to evaluate. The two models are essentially programmed to try to outdo each other. The generator is trying to get its output past the discriminator, while the discriminator is working to spot artificially generated data. This structured contest can optimize AI outputs to be more realistic without the need for additional human intervention. Though GANs work best for narrower applications (such as producing realistic photos or videos), rather than general purpose AI. Hallucination is the AI industry’s preferred term for AI models making stuff up – literally generating information that is incorrect. Obviously, it’s a huge problem for AI quality. Hallucinations produce GenAI outputs that can be misleading and could even lead to real-life risks — with potentially dangerous consequences (think of a health query that returns harmful medical advice). The problem of AIs fabricating information is thought to arise as a consequence of gaps in training data. Hallucinations are contributing to a push toward increasingly specialized and/or vertical AI models — i.e. domain-specific AIs that require narrower expertise – as a way to reduce the likelihood of knowledge gaps and shrink disinformation risks. Inference is the process of running an AI model. It’s setting a model loose to make predictions or draw conclusions from previously seen data. To be clear, inference can’t happen without training; a model must learn patterns in a set of data before it can effectively extrapolate from this training data. Many types of hardware can perform inference, ranging from smartphone processors to beefy GPUs to custom-designed AI accelerators. But not all of them can run models equally well. Very large models would take ages to make predictions on, say, a laptop versus a cloud server with high-end AI chips. [See:Training] Large language models, or LLMs, are the AI models used by popular AI assistants, such asChatGPT,Claude,Google’s Gemini,Meta’s AI Llama,Microsoft Copilot, orMistral’s Le Chat. When you chat with an AI assistant, you interact with a large language model that processes your request directly or with the help of different available tools, such as web browsing or code interpreters. LLMs are deep neural networks made of billions of numerical parameters (or weights, see below) that learn the relationships between words and phrases and create a representation of language, a sort of multidimensional map of words. These models are created from encoding the patterns they find in billions of books, articles, and transcripts. When you prompt an LLM, the model generates the most likely pattern that fits the prompt. (See:Neural network) Memory cache refers to an important process that boosts inference (which is the process by which AI works to generate a response to a user’s query). In essence, caching is an optimization technique, designed to make inference more efficient. AI is obviously driven by high-octane mathematical calculations and every time those calculations are made, they use up more power. Caching is designed to cut down on the number of calculations a model might have to run by saving particular calculations for future user queries and operations. There are different kinds of memory caching, although one of the more well-known isKV (or key value) caching. KV caching works in transformer-based models, and increases efficiency, driving faster results by reducing the amount of time (and algorithmic labor) it takes to generate answers to user questions. (See:Inference) A neural network refers to the multi-layered algorithmic structure that underpins deep learning — and, more broadly, the whole boom in generative AI tools following the emergence of large language models. Although the idea of taking inspiration from the densely interconnected pathways of the human brain as a design structure for data processing algorithms dates all the way back to the 1940s, it was the much more recent rise of graphical processing hardware (GPUs) — via the video game industry — that really unlocked the power of this theory. These chips proved well suited to training algorithms with many more layers than was possible in earlier epochs — enabling neural network-based AI systems to achieve far better performance across many domains, including voice recognition, autonomous navigation, and drug discovery. (See:Large language model [LLM]) Open source refers to software — or, increasingly, AI models — where the underlying code is made publicly available for anyone to use, inspect, or modify. In the AI world, Meta’s Llama family of models is a prominent example; Linux is the famous historical parallel in operating systems. Open source approaches allow researchers, developers, and companies around the world to build on top of one another’s work, accelerating progress and enabling independent safety audits that closed systems cannot easily provide. Closed source means the code is private — you can use the product but not see how it works, as is the case with OpenAI’s GPT models — a distinction that has become one of the defining debates in the AI industry. Parallelization means doing many things at the same time instead of one after another — like having 10 employees working on different parts of a project at the same time instead of one employee doing everything sequentially. In AI, parallelization is fundamental to both training and inference: modern GPUs are specifically designed to perform thousands of calculations in parallel, which is a big reason why they became the hardware backbone of the industry. As AI systems grow more complex and models grow larger, the ability to parallelize work across many chips and many machines has become one of the most important factors in determining how quickly and cost-effectively models can be built and deployed. Research into better parallelization strategies is now a field of study in its own right. RAMageddon is the fun new term for a not-so-fun trend that is sweeping the tech industry: an ever-increasing shortage of random access memory, or RAM chips, which power pretty much all the tech products we use in our daily lives. As the AI industry has blossomed, the biggest tech companies and AI labs — all vying to have the most powerful and efficient AI — are buying so much RAM to power their data centers that there’s not much left for the rest of us. And that supply bottleneck means that what’s left is getting more and more expensive. That includes industries like gaming (where major companies have had toraise prices on consolesbecause it’s harder to find memory chips for their devices), consumer electronics (where memory shortage could causethe biggest dip in smartphone shipmentsin more than a decade), and general enterprise computing (because those companies can’t get enough RAM for their own data centers). The surge in prices is only expected to stop after the dreaded shortage ends but, unfortunately, there’snot really much of a signthat’s going to happen anytime soon. Reinforcement learning is a way of training AI where a system learns by trying things and receiving rewards for correct answers — like training your beloved pet with treats, except the “pet” in this scenario is a neural network and the “treat” is a mathematical signal indicating success. Unlike supervised learning, where a model is trained on a fixed dataset of labeled examples, reinforcement learning lets a model explore its environment, take actions, and continuously update its behavior based on the feedback it receives. This approach has proven especially powerful for training AI to play games, control robots, and, more recently, sharpen the reasoning ability of large language models. Techniques like reinforcement learning from human feedback, or RLHF, are now central to how leading AI labs fine-tune their models to be more helpful, accurate, and safe. When it comes to human-machine communication, there are some obvious challenges — people communicate using human language, while AI programs execute tasks through complex algorithmic processes informed by data. Tokens bridge that gap: they are the basic building blocks of human-AI communication, representing discrete segments of data that have been processed or produced by an LLM. They are created through a process called tokenization, which breaks down raw text into bite-sized units a language model can digest, similar to how a compiler translates human language into binary code a computer can understand. In enterprise settings, tokens also determine cost — most AI companies charge for LLM usage on a per-token basis, meaning the more a business uses, the more it pays. So again, tokens are the small chunks of text — often parts of words rather than whole ones — that AI language models break language into before processing it; they are roughly analogous to “words” for the purposes of understanding AI workloads. Throughput refers to how much can be processed in a given period of time, so token throughput is essentially a measure of how much AI work a system can handle at once. High token throughput is a key goal for AI infrastructure teams, since it determines how many users a model can serve simultaneously and how quickly each of them receives a response. AI researcher Andrej Karpathy has described feeling anxious when his AI subscriptions sit idle — echoing the feeling he had as a grad student when expensive computer hardware wasn’t being fully utilized — a sentiment that captures why maximizing token throughput has become something of an obsession in the field. Developing machine learning AIs involves a process known as training. In simple terms, this refers to data being fed in in order that the model can learn from patterns and generate useful outputs. Essentially, it’s the process of the system responding to characteristics in the data that enables it to adapt outputs towards a sought-for goal — whether that’s identifying images of cats or producing a haiku on demand. Training can be expensive because it requireslotsof inputs, and the volumes required have been trending upwards — which is why hybrid approaches, such as fine-tuning a rules-based AI with targeted data, can help manage costs without starting entirely from scratch. [See:Inference] A technique where a previously trained AI model is used as the starting point for developing a new model for a different but typically related task – allowing knowledge gained in previous training cycles to be reapplied. Transfer learning can drive efficiency savings by shortcutting model development. It can also be useful when data for the task that the model is being developed for is somewhat limited. But it’s important to note that the approach has limitations. Models that rely on transfer learning to gain generalized capabilities will likely require training on additional data in order to perform well in their domain of focus (See:Fine tuning) Weights are core to AI training, as they determine how much importance (or weight) is given to different features (or input variables) in the data used for training the system — thereby shaping the AI model’s output. Put another way, weights are numerical parameters that define what’s most salient in a dataset for the given training task. They achieve their function by applying multiplication to inputs. Model training typically begins with weights that are randomly assigned, but as the process unfolds, the weights adjust as the model seeks to arrive at an output that more closely matches the target. For example, an AI model for predicting housing prices that’s trained on historical real estate data for a target location could include weights for features such as the number of bedrooms and bathrooms, whether a property is detached or semi-detached, whether it has parking, a garage, and so on. Ultimately, the weights the model attaches to each of these inputs reflect how much they influence the value of a property, based on the given dataset. Validation loss is a number that tells you how well an AI model is learning during training — and lower is better. Researchers track it closely as a kind of real-time report card, using it to decide when to stop training, when to adjust hyperparameters, or whether to investigate a potential problem. One of the key concerns it helps flag is overfitting, a condition in which a model memorizes its training data rather than truly learning patterns it can generalize to new situations. Think of it as the difference between a student who genuinely understands the material and one who simply memorized last year’s exam — validation loss helps reveal which one your model is becoming. This article is updated regularly with new information.
View
Nvidia has already committed $40B to equity AI deals this year
Nvidia continues to be a major investor in the AI ecosystem, committing more than $40 billion to equity investments in AI companies — and that’s just in these early months of 2026,according to CNBC. Much of that total comes from a single bet,a $30 billion investment in OpenAI. But CNBC reports that the chipmaker has also announced seven multi-billion dollar investments in publicly traded companies, most recently deals to invest up to $3.2 billion in glassmaker Corning and up to $2.1 billion in data center operator IREN. We’ve previouslyrounded up Nvidia’s investments in AI startups, including 67 venture deals in 2025. And according to FactSet data, it’s already participated in around two dozen investment rounds in private startups in 2026. The fact that Nvidia has been investing in some of its own customers has led to the recurring criticism that these arecircular dealsmoving money back-and-forth between the same companies. Wedbush Securities analyst Matthew Bryson said Nvidia’s investments fall “squarely into the circular investment theme,” but suggested that if successful, they could help the company build a “competitive moat.”
View
Databricks Has a ‘Too Much Opportunity Problem’ in India
For companies like Databricks, the country is becoming the next frontier of scale, experimentation, and eventually, product definition.
View
Laid-off Oracle workers tried to negotiate better severance. Oracle said no.
As was widely reported, Oracleaxed an estimated 20,000 to 30,000 peoplevia email on March 31. One of the employees cut that day told TechCrunch about the experience: “I had, like, this weird feeling in my stomach. I went to go sign into the VPN, and the VPN was like, ‘this user doesn’t exist anymore.’ Then I called my friend, and I was like, ‘Hey, can you see me in Slack?’ And she said, ‘No, your account’s been deactivated.’” The person soon received an email stating their role was terminated immediately. The severance offer arrived a few days later. But Oracle’s terms would quickly become a point of contention — and some laid-off employees would push back. Oracle offered fairly standard Corporate America terms to laid off employees. In exchange for signing a release waiving their right to sue, employees received four weeks of pay for the first year, plus one additional week per year of service, capped at 26 weeks. The company was also paying for one month of COBRA insurance. The catch: Although stock compensation often makes up a good chunk of a tech worker’s pay, particularly at Oracle, the company did not accelerate soon-to-vest RSUs. Any shares that hadn’t vested by the termination date were forfeited. That held true even for stock granted as retention incentives or in place of salary increases tied to promotions. One long-tenured employee lost $1 million in stock that was just four months from vesting; RSUs made up about 70% of his compensation,Time reported. Some employees also discovered that if they were classified as remote workers by the company, and didn’t work in a state with stronger worker provisions like California or New York, the company said they didn’t qualify for WARN Act protections. TheWARN Act is a lawthat requires companies conducting mass layoffs to give employees two months notice prior to letting them go. It’s triggered when 50 or more people are impacted at one location. By classifying employees as remote workers, the minimum location requirements can be sidestepped. Some people were unaware they were classified as remote workers, because they were near an office and worked on a hybrid schedule. Even if they were covered by the WARN Act, this did not necessarily extend severance, the former Oracle employee said. That’s because Oracle included the two-months’ WARN notice pay in its existing calculation of four-weeks, plus one week per year. For a short time, a group of employees tried to negotiate en masse with Oracle, according to a letter seen by TechCrunch. At least90 people signed a public petitionurging the database and cloud computing giant to match the terms of other big tech companies conducting mass layoffs in the name of AI. For instance, Meta’s severance package, according to an email published by Business Insider, started at 16 weeks of base pay, plus two weeks for every year of employment and covered COBRA for 18 months. Microsoft, which extended voluntary retirement offers to long-serving employees, provided accelerated stock vesting, a minimum of eight weeks’ pay, and an additional one to two weeks for every six months of service, depending on rank, theSeattle Times reported. And Cloudflare, which just cut 20% of its employees,offered lump sum severancethat was the equivalent of base pay through the end of 2026, plus healthcare coverage through the end of the year, and accelerated vesting of stock through August 15. So if an employee was close to obtaining another tranche, they will get it. Oracle declined to negotiate, according to an email seen by TechCrunch. It was a take-it-or-leave scenario, the employee said. When asked about its severance terms, classifying employees as remote, and the failed attempt by employees to negotiate more, Oracle declined to comment. Such a reaction from the company isn’t a surprise, not even to those who hoped to negotiate. But it does underscore that for all the theoretical high pay (often via stocks) and perks that tech workers enjoy when it’s an employees’ market, they have very few protections in place when it isn’t.
View
Cloudflare says AI made 1,100 jobs obsolete, even as revenue hit a record high
Cloudflare on Thursday joined a growing list of tech companies — including Meta, Microsoft, and Amazon — that have reported increased revenue alongside massive layoffs, attributing both trends to their use of AI. Cloudflare, which provides internet security and performance services to millions of websites worldwide, announced it was cutting its workforce by approximately 20%, which equates to 1,100 people, it said as part of its first quarter 2026 earnings report on Thursday. “We’ve never done something like this in Cloudflare’s history,” co-founder and CEO Matthew PrincesaidThursday on the quarterly conference call, marking the first mass layoff in the company’s 16-year history. The company is cutting people from all teams and geographies except for salespeople who carry revenue quotas, CFO Thomas Seifert detailed on the call. The news of the workforce cuts came as the companyreportedquarterly revenues of $639.8 million, a 34% year-over-year increase and the highest single quarter in the company’s history. However, this was coupled with a loss of $62.0 million compared with losing $53.2 million in the year-ago quarter. That widening loss, even as revenue surged, highlights a familiar paradox in Cloudflare’s story: the company is growing fast but has yet to turn a consistent profit. But the loss was a smaller percentage of revenue, and the quarter was coupled with a lot of other positive indicators. For instance, Cloudflare reported that it had over $2.5 billion in “remaining performance obligations,” a year-over-year growth of 34%. RPO is the favorite metric these days to indicate revenue under contract but not yet delivered. Hence, Prince insisted, the 20% cuts were not to reduce expenses but were strictly because of its use of AI. “Today’s actions are not a cost-cutting exercise or an assessment of individuals’ performance; they are about Cloudflare defining how a world-class, high-growth company operates and creates value in the agentic AI era,” Prince and Cloudflare co-founder and president, Michelle Zatlyn,wrotein a related blog post about the layoffs. Prince acknowledged on the call that even though Cloudflare has been selling AI-powered products, it was at first cautious about adopting AI itself. “Internally, the tipping point was last November. At that point, across our teams, we began to see massive productivity gains, team members that were two, 10, even 100 times more productive than they had been before. It was like going from a manual to an electric screwdriver,” he described. “Cloudflare’s usage of AI has increased by more than 600% in the last three months alone,” he added. Prince highlighted the internal use of AI coding, saying that virtually the entire R&D team is now using the company’s own Workers platform — a tool that lets developers build and run software directly on Cloudflare’s global network — including its vibe coding feature. He also noted that 100% of the code produced this way and deployed for use in Cloudflare’s products is “now reviewed by autonomous AI agents.” But it’s not just developers who are using AI internally, he said. “Employees across the company from engineering to HR to finance to marketing run thousands of AI agent sessions each day to get their work done.” As a result, these highly productive, AI-powered employees require fewer support staff, he argued. “A lot of the support people that provide support behind them, those roles aren’t going to be the roles that, you know, drive companies going forward,” Prince said. Interestingly, Prince says that Cloudflare “will continue to hire people, and we’ll continue to invest in them because the people that are embracing these tools are just so much more productive than we’d ever seen before. I would guess that in 2027 we’ll have more employees than we did at any point in 2026.” Cloudflare said it ended its first quarter before layoffs with a headcount of about 5,500. The pattern Prince described — deploying AI gains as justification for workforce reductions even during a period of strong revenue growth — is fast becoming a familiar script across the tech industry. Whether it reflects true structural transformation or acts as convenient cover for cost discipline is a question that investors and employees will be wrestling with for some time to come. When asked by an analyst on the call why the company needed to cut so deeply after such a good quarter, Prince said, “Just because you’re fit doesn’t mean you can’t get fitter.”
View
Intel’s comeback story is even wilder than it seems
Bloomberg has adeep divethis week into how Intel CEO Lip-Bu Tan is trying to rescue one of Silicon Valley’s most storied, and stumbling, chipmakers. It’s worth a read, but it actually undersells the most jaw-dropping part of the story: Intel’s stock has risen a stunning 490% over the past year, a bet by Wall Street that may be running well ahead of the company’s actual turnaround. Tan, who took over inMarch of last year, has spent much of his first year schmoozing rather than restructuring — locking in asweetheart dealwith the U.S. government (now Intel’s third-largest shareholder), cozying up to Elon Musk on afactory partnership, and reportedly landing preliminary manufacturing agreements with both Apple and Tesla. The fundamentals are still messy. Intel’s chip yields lag well behind industry leader TSMC, and employees tell Bloomberg that Tan has been light on specifics internally, with some teams adjusting missed deadlines rather than recovering from them. But investors are betting big on the bigger picture. Whether the execution follows is the multi-billion-dollar question.
View
Last 24 hours to get 50% off a second pass to TechCrunch Disrupt 2026
Today is the last day. At 11:59 p.m. PT, the 50% off second pass offer forTechCrunch Disrupt 2026ends. After that, prices go up, and the option to bring a partner, co-founder, or colleague with you at half the cost disappears. Register now to lock in your savings.Save up to $410 on your pass and get 50% on a second pass. Disruptisn’t a single-track experience. It’s multiple conversations happening at once. Sessions overlap. Introductions lead to something else an hour later. Patterns only become clear after you’ve seen the same idea from different angles. When you go alone, you see only part of it. When you bring someone, you see more, and more importantly, you understand more. You compare notes in real time, challenge assumptions, and make decisions while the context is still fresh.Get a discounted second pass now. You and your plus-one will have access to: That’s not a small difference. It’s the difference between leaving with ideas and leaving with direction for your next steps. And after tonight, that second perspective costs more. This is your last day to save 50% on a second pass.Choose your tickets. From October 13–15 in San Francisco atDisrupt,the startup world will be in the same place at the same time, turning conversations into capital, ideas into companies, and connections into trajectories. They’ll be trading signals, testing assumptions, and deciding what matters based on what they’re seeing in real time. When you act now to secure your pass —and a second at 50% off— you’ll be in the room while those decisions (and discussions) are taking shape. Across 250+ sessions, you’ll explore real-world playbooks (not theory), covering: Those conversations don’t pause when the event ends. They carry forward into follow-ups, deals, partnerships, and decisions made in the weeks that follow. If you’re not there, you’re not just missing the event. You’re reacting later to conclusions other people reached sooner.Buy a pass to Disrupt today and get a second one for 50%off to be a part of the conversations. Knowing you have a strong idea isn’t enough. You need clarity on where to take it, who to partner with, and how to fund it. Without that clarity, decisions stall. Roadmaps stretch. Opportunities sit just long enough to lose momentum. Disruptcompresses that uncertainty. You see how decisions get made — onstage, in roundtables, and in conversations that build on each other over three days. Better outcomes come from: Miss that window, and you’re back to piecing together secondhand insight, slower feedback loops, and decisions made without the same level of context. This is your final day to get a second pass for 50% off.Register now before prices increase at 11:59 p.m. ET tonight. After tonight, you can still attendDisrupt. But you’re more likely to go alone — and that changes the experience. It means choosing between sessions instead of covering more ground. Processing everything yourself instead of testing it in real time. Following up later instead of leaving with shared clarity. That’s the real cost. Not just paying more, but also getting less out of being there.Lock in your 50% savings on a second ticketto show up more intentionally. Only hours remain to buy a pass toDisruptand get a second for 50% off. The offer ends tonight at 11:59 p.m. PT. Right now, you can still: After today, that advantage is gone. Buy one pass to Disrupt andget 50% off the second of the same ticket type. Decide who you’re bringing — and secure your passes before midnight tonight. Because missing this isn’t just about price. It’s about showing up with less context, less coverage, and less clarity than the people who didn’t wait.
View
The “people’s airline” and the enterprise AI gold rush
Everyone wants a piece of the enterprise AI pie, and this week, we saw a string of companies making their moves. FromAnthropic and OpenAI announcing new joint venturestargeting enterprise AI deployment toSAP dropping $1Bon German AI startup Prior Labs, it’s becoming clear that if you’re a startup building enterprise tools, you’re likely an acquisition target. On this episode of TechCrunch’sEquitypodcast, hosts Kirsten Korosec, Anthony Ha, and Sean O’Kane dig into the week’s enterprise AI deals, thexAI-Anthropic compute arrangement, and what it all means ahead of what could be a big IPO season. Listen to the full episode to hear about: Subscribe to Equity onYouTube,Apple Podcasts,Overcast,Spotifyand all the casts. You also can follow Equity onXandThreads, at @EquityPod.
View